OptiGein Design Docs
Home › Which design do I need?

Which design do I need?

OptiGein Design can build sixteen kinds of design. That sounds like a lot, but the choice is usually easy once you know which stage of the experiment you are at. This page walks you through the three stages, points each one at the right family, and lists every design in a single table so you can compare them at a glance.

Three stages, not sixteen decisions

Almost every study moves through the same three stages. You do not always run all three, but knowing where you are is the fastest way to narrow the list.

Screening. You have many candidate factors and no idea yet which ones matter. The goal is to test them all in as few runs as you can afford and find the vital few that move the response. You accept that you cannot separate every interaction at this stage; you just want to know which factors are worth studying further. Screening designs live in the Factorial family, with one useful screen (Definitive Screening) in the Response Surface family.

Characterisation. You are down to a handful of important factors and now you want to understand how they work together, including which pairs interact. This is what full and fractional two-level designs are built for, so you stay mostly in the Factorial family.

Optimisation. You know the factors that matter and you want the settings that give the best result. That means fitting curvature, not just straight-line effects, so each factor needs more than two settings. This is the job of the Response Surface family.

The New Design screen groups the catalogue into these families, so once you have named your stage you only have a few entries to read.

The New Design catalogue with the four design families in a tree on the left
The catalogue, grouped into Factorial, Response Surface, Mixture and Custom Designs. Selecting an entry shows what it is for and how many runs it costs.

Mixture designs are a special case

There is a fourth situation that does not fit the stages above. If your factors are the ingredients of a recipe, they are not independent: they have to add up to a fixed total. Push one ingredient up and the others must come down to keep the batch at one hundred percent. An ordinary factorial assumes you can set each factor on its own, so it simply does not apply here. That is why mixtures get their own family.

How to tell. If changing one factor forces the others to change so the total stays constant, you have a mixture. Ratios of ingredients, percentages of a formulation and parts of a blend are all mixtures. Use the Mixture family. If a recipe and process settings both matter in the same study, use Optimal (Combined) under Custom Designs.

The full catalogue

Every design the software builds is below, with its family, how many factors it accepts, what it is for, and a rough run count. Treat the run counts as a guide; the builder shows you the exact number before it creates anything, and options like centre points and replicates change it.

Design Family Factors What it is for Rough runs
Regular Two-LevelFactorial2 to 21 Screen or characterise with every factor at a low and a high setting 2 to the power of the factor count for a full set, fewer for a fraction
Multilevel CategoricFactorial1 to 12 Cross every named level of every categoric factor The level counts multiplied together, so it grows quickly
Plackett-BurmanFactorial2 to 23 Screen many factors for main effects only (Resolution III) 12, 20 or 24
Min-Run CharacterizeFactorial6 to 20 Keep every main effect and every two-factor interaction readable (Resolution V) 1 + k + k(k-1)/2, about 22 at six factors
Min-Run ScreenFactorial5 to 50 Trim a long factor list, main effects clear of two-factor interactions (Resolution IV) About two per factor, roughly 2k
Taguchi OAFactorial3 to 15 Classic orthogonal arrays for main effects Fixed by the array, 4 to 27
Central CompositeResponse Surface2 to 10 Map curvature and locate an optimum, five settings per factor Factorial core + 2k axial + centre points, 13 for two factors
Box-BehnkenResponse Surface3 to 5 Fit a quadratic model with no run at all extremes at once 2k(k-1) + centre points, about 17 at three factors
Optimal (Custom)Response Surface1 to 10 Smallest run list for a model you state, numeric and categoric factors welcome Model terms + 10 by default
Definitive ScreeningResponse Surface4 to 20 Compact three-level screen that also shows curvature About 2k + 1
Simplex LatticeMixture2 to 12 Spread blends evenly across the recipe space to a chosen polynomial degree Set by component count and lattice degree
Simplex CentroidMixture3 to 10 Every equal-parts blend, best when component synergy is the question 2 to the power of the component count, minus 1, plus axial blends if augmented
Optimal (Custom)Mixture2 to 10 Fewest blends that support a Scheffe linear or quadratic model Model terms + 10 by default
Optimal (Combined)Custom Designs2 to 8 One experiment for the recipe and the process together Crossed model terms + 10 by default
Blank SpreadsheetCustom Designs1 to 20 An empty run sheet you fill in by hand or by paste However many rows you set
Import Data SetCustom Designs1 to 30 Bring in results you already have and go straight to analysis Matches the rows you paste

In the run counts, k is the number of factors. The worked example used throughout this documentation is a face-centred central composite design on two numeric factors, A: HPMC E15 (2 to 4 % w/v) and B: Glycerol Ester (0.5 to 1.5 % w/v). Its run count follows the Central Composite row: four factorial corners, four axial runs and five centre points give 13 runs in all.

Read the panel. Click any entry in the catalogue and the panel on the right tells you what that design does and, once you set the factor count and options, exactly how many runs it will cost. Nothing is built until you press Finish, so you can compare a few candidates freely before you commit.

Let the wizard choose for you

If you would rather answer a few questions than read the whole catalogue, open the Design Wizard from the Design Tools menu. It asks three things: what you are trying to do (screen, characterise, optimise, or just analyse data you already have), whether your factors are independent process settings or mixture components, and how many factors and runs you have in mind. It then lists a short set of suitable designs with an estimated run count for each, and marks any that would exceed your run budget.

The final wizard step listing suggested designs with estimated run counts
The wizard's last step. Set your factor count and run budget, then pick from the suggestions. Finish opens the builder on that design with the factor count already filled in.

The wizard is a good starting point when the choice is not obvious, or when you want a quick sanity check on how many runs an idea will cost. When you know exactly what you want, going straight to the catalogue is faster.

Go deeper on a family

Once you have a family in mind, its own page explains every option and what it does to the run sheet.